Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AlphaGo beats Lee Sedol 3-0 [video]

youtube.com

Post reply on HN
Print view View original

321–330 of 428 posts

Re: AlphaGo beats Lee Sedol 3-0 [video]

#321

Post by hacker42 »

suryabhupa wrote: post #318
YeGoblynQueenne wrote: post #314

Earlier quoted context omitted.

I don't know whether this is true. If AlphaGo could do just as well without search, then why did it use search at all? But in any case, I'm not necessarily disputing that. I'm particularly refuting the claim that the AlphaGo architecture is identical to the one that learned to play Atari games and that Deepmind have advertised as a general game-playing agent. My comment here is specifically in reply to the GP who rep…

The idea is that even there's a policy network that is able to decide at some point what the best possible move is, the tree search is done to refine this choice and to "evaluate" it. This is why a value network is derived from policy network and is used in conjunction with MCTS to make sure that the moves AlphaGo picks are good ones.

It is necessary to make multiple alternatives in the tree comparable in an easy way (and nothing is better comparable than scalars). They could also go about training a network that compares two positions to decide which one is superior, but that would require much more computation. Or another alternative would possibly to learn the value somehow jointly with the action selection, but that would possibly also be harder both to train and evaluate.

Re: AlphaGo beats Lee Sedol 3-0 [video]

#322

Post by skybrian »

Eliezer wrote: post #79

My (long) commentary here: https://www.facebook.com/yudkowsky/posts/10154018209759228 Sample: At this point it seems likely that Sedol is actually far outclassed by a superhuman player. The suspicion is that since AlphaGo plays purely for probability of long-term victory rather than playing for points, the fight against Sedol generates boards that can falsely appear to a human to be balanced even as Sedol's probabili…

On the other hand, in many ways AlphaGo plays much like a human professional. The openings and fights were not radically different. The professional commentators did not have a hard time explaining why most moves were good. The structure of the board was understandable. Some moves were "forced" moves that were predictable and many others were good moves that a human would play. The human player isn't efficient enough to win, but attacks and defenses still mostly work as intended.

This suggests that Go space was already very well understood and there aren't radically different play styles we've somehow overlooked. AI is not a magic wand that radically changes how the game works.

However there may be some bias due to AlphaGo having trained on human games and playing against a human. The real proof will happen when they redo the retraining from scratch.

Re: AlphaGo beats Lee Sedol 3-0 [video]

#323

Post by sbierwagen »

ionforce wrote: post #296
Eliezer wrote: post #79

My (long) commentary here: https://www.facebook.com/yudkowsky/posts/10154018209759228 Sample: At this point it seems likely that Sedol is actually far outclassed by a superhuman player. The suspicion is that since AlphaGo plays purely for probability of long-term victory rather than playing for points, the fight against Sedol generates boards that can falsely appear to a human to be balanced even as Sedol's probabili…

> AI alignment theory What is this?

The problem of making sure an AI does what you want it to.

http://lesswrong.com/lw/n4l/safety_engineering_target_select...

Re: AlphaGo beats Lee Sedol 3-0 [video]

#324

Post by technolem »

argonaut wrote: post #247
anonaoeu wrote: post #103

Earlier quoted context omitted.

If an AI has any motivation at all, say, to make paperclips as efficiently as possible, then any threat to its existence is a threat to its objective function - namely, to create paperclips. A hyper-intelligent entity who is instructed to optimize for paperclips created will therefore proactively remove threats to its existence (i.e. its paperclip-creating functionality) and might possibly turn the entire solar syste…

Such an entity would not be hyper-intelligent. It would be idiotic. One huge hole for me in the paperclip argument is that an AI capable of that kind of power would not be stupid enough to misinterpret a command - it would be intelligent enough to infer human desires.

Of course it would. But, it's not programmed to care about what you meant to say. It will gladly do what it was mis-programmed to do instead. You can already see this kind of trait in humans, where instinct is mis-aligned with intended result. Such as procreation for fun + birth control.

Re: AlphaGo beats Lee Sedol 3-0 [video]

#325

Post by hacker42 »

YeGoblynQueenne wrote: post #299
Eliezer wrote: post #79

My (long) commentary here: https://www.facebook.com/yudkowsky/posts/10154018209759228 Sample: At this point it seems likely that Sedol is actually far outclassed by a superhuman player. The suspicion is that since AlphaGo plays purely for probability of long-term victory rather than playing for points, the fight against Sedol generates boards that can falsely appear to a human to be balanced even as Sedol's probabili…

>> But Go is rich enough to demonstrate strong cognitive uncontainability on a small scale. In a rich and complicated domain whose rules aren't fully known, we should expect even more magic from superhuman reasoning - solutions that are better than the best solution we could imagine, operating by causal pathways we wouldn't be able to foresee even if we were told the AI's exact actions. Hang on. Where are we going to…

> What are we going to train such a learner with, if we ourselves don't fully know the rules of the domain, as you point out?

This does not make any sense if you assume general intelligence. A physicist who makes a discovery also was not trained to know about this new law or rule beforehand (how could he?).

Re: AlphaGo beats Lee Sedol 3-0 [video]

#326

Post by orangecat »

21 wrote: post #58
badminton1 wrote: post #49

Earlier quoted context omitted.

It is not a tool if you can't control it. Politicians fool countries delivering empty promises about better health, education and security. A supraintelligent AI could promise making humans rich, healthy and powerful, to then break its promise and dominate the world. https://en.wikipedia.org/wiki/AI_box

Do you feel guilty when you break your promise against your cat? Do you even think for a nanosecond if it's ethical to lie to it? Of course, a cat is not conscious. But compared to an AI, we might also be considered pretty low consciousness beings, or at least beings in front of which you don't justify yourself.

Do you feel guilty when you break your promise against your cat?

If some unforeseen event occurred and I had to abandon my cat, thereby breaking my promise that I would take care of her, I would definitely feel guilty about it.

Of course, a cat is not conscious.

Either this is a nonstandard definition of "conscious", or you haven't met many cats.

Re: AlphaGo beats Lee Sedol 3-0 [video]

#327

Post by sbierwagen »

argonaut wrote: post #247
anonaoeu wrote: post #103

Earlier quoted context omitted.

If an AI has any motivation at all, say, to make paperclips as efficiently as possible, then any threat to its existence is a threat to its objective function - namely, to create paperclips. A hyper-intelligent entity who is instructed to optimize for paperclips created will therefore proactively remove threats to its existence (i.e. its paperclip-creating functionality) and might possibly turn the entire solar syste…

Such an entity would not be hyper-intelligent. It would be idiotic. One huge hole for me in the paperclip argument is that an AI capable of that kind of power would not be stupid enough to misinterpret a command - it would be intelligent enough to infer human desires.

Yeah, but why would it want to? I can perfectly infer the values of an earthworm, but I don't dedicate all my resources to making worms happy.

Re: AlphaGo beats Lee Sedol 3-0 [video]

#329

Post by orangecat »

ap22213 wrote: post #147
pushrax wrote: post #31

It's important to remember that this is an accomplishment of humanity, not a defeat. By constructing this AI, we are simply creating another tool for advancing our state of being. (or something like that)

I hate to be cynical, but I'm sure many of my ancestors were also told to believe similar things. I know for certain that my grandparents and great grandparents believed that technology would create such progress that people of my generation would not have to work, and all would have leisure time. AI is more likely to evolve into a tool to be used by the few to control the many.

We do in fact have more leisure time: https://www.stlouisfed.org/publications/regional-economist/j.... And that doesn't count reading Facebook at work.

Re: AlphaGo beats Lee Sedol 3-0 [video]

#330

Post by adrianN »

skybrian wrote: post #322
Eliezer wrote: post #79

My (long) commentary here: https://www.facebook.com/yudkowsky/posts/10154018209759228 Sample: At this point it seems likely that Sedol is actually far outclassed by a superhuman player. The suspicion is that since AlphaGo plays purely for probability of long-term victory rather than playing for points, the fight against Sedol generates boards that can falsely appear to a human to be balanced even as Sedol's probabili…

On the other hand, in many ways AlphaGo plays much like a human professional. The openings and fights were not radically different. The professional commentators did not have a hard time explaining why most moves were good. The structure of the board was understandable. Some moves were "forced" moves that were predictable and many others were good moves that a human would play. The human player isn't efficient enough…

AlphaGo learned the game from humans by looking at recorded games. It's not suprising that its style mimicks its master's.
Post reply on HN